Cursor 2.0 Unveils Multi-Agent AI Workflows and New Composer Model

Cursor has released version 2.0, a significant update introducing multi-agent capabilities and a new proprietary large language model, Composer. This multi-agent system, drawing parallels to Cloud Code’s sub-agents, allows developers to execute multiple AI agents concurrently, even on the same task, through a new dedicated interface. Composer, Cursor’s in-house model, is positioned as a direct competitor to Anthropic’s Sonnet, expanding Cursor’s existing support for models like GPT, Grok, DeepSeek, and Cloud Code. Initial metrics for Composer are described as generic, prompting users to gauge its performance through practical application.

Practical evaluations of Composer indicate generally fast response times, though potentially slightly slower than Cloud Code for some tasks. While Composer can generate functional applications with animations and data seeding, the resulting code often requires manual correction, debugs various errors, and occasionally utilizes outdated packages, underscoring that AI-generated code still necessitates human oversight. The multi-agent interface demonstrates potential for parallel task execution, particularly for distinct coding tasks like integrating Stripe or adding end-to-end testing, without immediate evidence of interference. However, running multiple agents simultaneously significantly increases resource consumption, potentially impacting editor performance. Cursor 2.0 also integrates browser functionality, allowing AI agents to interact with running applications, select UI elements, and process frontend console errors, enhancing AI-assisted debugging and UI modification workflows.